MTBF vs MTTF: the difference, when each applies, and how both mislead
The short version, if that's all you need:
- MTTF — Mean Time To Failure — applies to things you replace: bearings, seals, lamps, fuses. It's the expected lifetime of one unit.
- MTBF — Mean Time Between Failures — applies to things you repair: pumps, compressors, servers. It's the average operating time between successive failures of the same system.
- The divider is one question: after the failure, is it the same item (repaired) or a new one (replaced)?
That answer is correct and will pass any exam. The rest of this page is the part that actually matters in practice: how each is computed, why the two get conflated, and the three ways these numbers routinely mislead the people they're reported to.
MTTF: the mean of a life distribution
For a non-repairable item, life is a random variable with some distribution — Weibull, Lognormal, whatever fits. MTTF is that distribution's mean:
MTTF = ∫₀^∞ R(t) dt (the area under the reliability curve)
The right way to estimate it is to fit the life distribution to your data — including the units that haven't failed, which is most of them in any maintained fleet (see censored data and suspensions for how badly it goes wrong otherwise) — and take the fitted mean.
MTBF: a rate, dressed as a time
For a repairable system, failures form a sequence in time. In its common industrial use, MTBF is:
MTBF = total operating time / number of failures
Note what this is: the reciprocal of an average failure rate. It says nothing about the pattern of failures — a system failing like clockwork every 1,000 hours and a system alternating 100-hour and 1,900-hour intervals both report MTBF = 1,000 hours. One of them is telling you something is wrong with your overhauls; the average can't hear it. (Trend analysis of repairable systems — is it degrading or improving? — is a Crow-AMSAA question, not an MTBF question.)
The three standard deceptions
1. An MTBF of 100,000 hours does not mean it lasts 11 years. MTBF figures on datasheets describe the failure rate during useful life, before wear-out. A hard drive with a million-hour MTBF isn't expected to run for 114 years; it's expected to fail at a rate of about 1 per 114 drive-years while within its (say) 5-year design life. Fleet-level rate, not unit-level lifetime — most datasheet-MTBF outrage dissolves once this distinction lands.
2. Most units are dead before the mean. For the exponential distribution (constant failure rate — the assumption buried in most MTBF arithmetic), the probability of failing before the MTBF is 1 − 1/e ≈ 63.2%. For a wear-out mode (Weibull β = 2.5, η = 6,000 h) the mean is 5,324 hours and about 52% of units are dead by then. The mean is not "when it fails"; it's the balance point of a skewed distribution. If the question is "when should we intervene?", you want a B10 life, not a mean.
3. The mean hides the failure pattern — which decides your strategy. Two components, both MTTF = 5,000 hours. One has Weibull β = 0.9 (infant mortality): replacing it on a schedule increases failures, because replacements die young. The other has β = 3 (wear-out): scheduled replacement is exactly right. Identical means, opposite maintenance policies. The number everyone reports contains zero bits of the information the decision needed.
How to compute each honestly
| MTTF (non-repairable) | MTBF (repairable) | |
|---|---|---|
| Data | Lifetimes of units, including suspensions | Operating hours and failure count per system |
| Method | Fit a life distribution (MLE), take its mean | Total time / failures — after checking the rate is stable |
| Watch out for | Failures-only averaging (biases low, often 2×) | Trends: an improving or degrading system makes one number meaningless |
| Better companions | B10 life, full distribution, confidence bounds | Rate over time, Crow-AMSAA trend, Duane plot |
Both columns share one prerequisite people skip: the arithmetic is only as good as the event data. "Total time / failures" silently assumes an exponential world (constant rate). It's a fine screening number — it's just not a lifetime, a guarantee, or a maintenance interval.
Frequently asked questions
Is MTBF the same as MTTF plus repair time?
In some standards MTBF is decomposed as MTBF = MTTF + MTTR (time between failures = time to failure + time to repair). In practice repair time is usually negligible against operating time, and industrial usage treats MTBF simply as mean operating time between failures. What matters is knowing which convention a number in front of you used.
Can I convert MTBF to failure rate?
Under a constant-failure-rate (exponential) assumption, failure rate λ = 1/MTBF. An MTBF of 50,000 hours is a rate of 2×10⁻⁵ failures per hour. The conversion is only as valid as the constant-rate assumption — it fails for wear-out and infant-mortality patterns.
What's the probability of surviving to the MTBF?
With a constant failure rate, R(MTBF) = 1/e ≈ 36.8% — nearly two-thirds of units fail before it. For wear-out modes the fraction failing before the mean is typically 50–60%. Either way, the mean is not a "safe until" time.
Which should I use for a maintenance interval?
Neither, directly. Intervals should come from the fitted life distribution — a B-life for risk-based limits, or a cost-optimal replacement calculation that weighs planned against unplanned replacement cost. A mean alone cannot answer "when", only "how often on average".